Quiz sur l'IA dans la lecture labiale et la reconnaissance visuelle de la parole

Les réponses et les explications apparaissent au fur et à mesure que vous répondez au quiz.

1. What ambiguity does a viseme grouping describe in visual speech recognition?
   • Every sound has a uniquely recoverable image in every frame
   • A camera directly measures a word's meaning
   • Different speech sounds can share similar visible mouth movements
   • The speaker's filename determines the phoneme sequence
2. How can a visual stream help an audio-visual recognizer when acoustic noise increases?
   • It guarantees correct recognition regardless of occlusion or lighting
   • It can supply additional visible articulation cues when the audio is degraded
   • It removes every need to evaluate the fusion model
   • It makes the acoustic signal physically noiseless
3. Qu’est-ce que le frontal convolutif 3D dans un modèle de lecture labiale aide à capturer ?
   • La couleur des yeux de l'orateur
   • La température ambiante
   • L'accent du locuteur à l'écrit
   • Modèles de mouvement à court terme sur des images consécutives
4. Which capture conditions can hide information needed by a visual-speech model?
   • Head turns, poor lighting, or an obstruction over the mouth
   • A clear frontal view as the only guaranteed failure condition
   • A rule that higher resolution removes all ambiguity
   • An audio file's compression setting with no effect on the video
5. What evaluation scope should be attached to the original LipNet sentence-recognition results?
   • Every spontaneous real-world conversation in every language
   • The constrained GRID benchmark used in the study
   • A guarantee that all professional lip readers are always outperformed
   • Only written text with no visual speech input
6. Une préoccupation sociétale majeure concernant l’IA de lecture labiale est la suivante :
   • Il consomme trop d'électricité
   • Il ne peut pas fonctionner sur des ordinateurs
   • Une utilisation secrète permet une surveillance sans consentement
   • Cela ne fonctionne que dans une seule langue
7. Why might a visual-speech pipeline crop a mouth-centered region before temporal modeling?
   • To focus processing on relevant articulation rather than unrelated image background
   • To remove the need for any information about movement
   • To guarantee that every phoneme becomes visually unique
   • To supply the final transcript directly from the crop coordinates
8. What alignment requirement does CTC avoid in the LipNet training setup?
   • All transcript-level training targets
   • All observations of the speaker's mouth
   • Every temporal dependency in the recognition problem
   • A supplied character label for every individual video frame